This study reexamines REPEAT COMPLAINT ADDRESS
POLICING: TWO FIELD EXPERIMENTS IN MINNEAPOLIS, 1985-1987 (ICPSR
9788), conducted by the Minneapolis Police Department and Lawrence W.
Sherman of the Crime Control Institute. Several features of the
original data distort the one-to-one correspondence between a 911 call
and an event, such as the occurrence of multiple versions of the same
call in the databases. The current study identifies and attempts to
correct these occurrences by applying multiple levels of data cleaning
procedures to establish a better one-to-one call-to-event
correspondence. Using this data collection, researchers should be able
to determine if the conclusions of the original study should be
modified after the distortions in the original data have been
The original RECAP study was designed as a
randomized controlled field experiment to study the strategy of
problem-oriented policing. It used individual geographic addresses in
the city of Minneapolis as the unit of analysis and compiled a
database from archived 911 computer tapes created by the Minneapolis
Emergency Communications Center (MECC). Addresses were then ranked
according to the frequency with which they appeared in the database.
Five hundred addresses were chosen for the experiment and were
randomly assigned to control and experimental groups, each consisting
of 125 commercial and 125 residential addresses. The current study
identifies three types of distortion in the original databases: True
Mirrors, Contemporaneous Calls, and Migrating Calls. True Mirrors are
multiple versions of a single call that are created by the
idiosyncrasies of the MECC computer system. Contemporaneous Calls are
multiple calls to MECC regarding a single event. Migrating Calls are
calls that are associated with one address, but then are reassigned to
another address, while the original address remains in the
database. The data in this study are recreations of the original RECAP
databases, which were then modified. (The actual databases of the
original study were not available, and consequently it was necessary
to reconstruct them.) The sixteen data files are organized such that
each file corresponds to either an "A", "B", "C", or "D" level
of data cleaning. Each file contains either commercial or residential
addresses from either the baseline or experimental years. The "A"
level data is a close approximation of the databases used in the
original RECAP study. The "B" level data is a subset of the "A"
level data created by deleting True Mirrors. The "C" level data are
a subset of the "B" level data without concurrent calls regarding a
single event. The "D" level data contain only dispatched calls to
The original RECAP study used random selection of
addresses. The current study modifies the random selection by the
application of data cleaning procedures.
The sample was drawn from a universe consisting of the
2,000 most frequently referenced addresses in calls to the Minneapolis
Emergency Communications Center (911) during 1986-1987.
Geographic addresses in the city of Minneapolis,
Minneapolis Emergency Communications Center 911
The sixteen data files use the same
variables. Each record contains information regarding a specific 911
call. Data were collected on the volume of calls to an address, the
date and time of the 911 call, the nature of the call (i.e., domestic
disturbance, traffic accident), and the disposition of the call.